What Are AI Usage Cards?

The Short Version

AI Usage Cards are a standardized way for researchers to document how they used AI tools in their work. Think of them as a structured disclosure form that captures which AI tools you used, what you used them for, how you verified the outputs, and what ethical considerations you addressed. You can generate one for free at ai-cards.org or use the LaTeX template on Overleaf.

The framework was developed by a research team at the University of Göttingen and published at the ACM/IEEE Joint Conference on Digital Libraries (JCDL 2023, DOI: 10.1109/JCDL57899.2023.00060). It was created in response to a simple problem. Researchers everywhere were starting to use AI tools like ChatGPT, Copilot, and Midjourney, but there was no agreed-upon way to report that usage. Everyone was improvising, which made it impossible to compare disclosures across papers or institutions.

Why Standardization Matters

Before AI Usage Cards, researchers who wanted to be transparent about their AI usage had to wing it. Some added a sentence to their acknowledgments. Others wrote a paragraph in their methods section. A few created custom appendices. The result was a patchwork of disclosures with no consistent structure, making it difficult for reviewers and readers to quickly find the information they needed.

AI Usage Cards solve this by giving everyone the same template. When a reviewer picks up a paper with an AI Usage Card attached, they know exactly where to look for which tools were used, what tasks those tools performed, and how the researcher validated the AI-generated outputs. This consistency is valuable for the same reason that standardized reference formats are valuable. It reduces friction and makes the information accessible.

The Four Sections of an AI Usage Card

Every AI Usage Card is organized into four main sections. Each one captures a different dimension of AI involvement in a research project.

1. Project Details

This section records the basic metadata about your project and your AI usage. It includes the names and affiliations of the authors, the AI models and tools that were used (with version numbers where available), the time period during which the tools were used, and the access method (API, web interface, local deployment).

Version information matters more than many researchers realize. GPT-4 in March 2024 and GPT-4 in January 2026 are not the same model. If someone wants to understand or reproduce your work, they need to know exactly which version of a tool you were working with. This section makes that information explicit.

2. Methodology and Experiments

Here you document how AI was involved in the research design and experimental process itself. This goes beyond writing assistance. Did you use an LLM to generate hypotheses? Did you use AI to help design your survey instruments? Did an AI tool suggest statistical methods or help you compare experimental approaches?

This section also covers AI involvement in data collection and analysis. If you used an AI model to code qualitative data, classify images, or preprocess text, this is where you record it. The goal is to give readers a clear picture of where human judgment drove the research process and where AI tools played a supporting role.

For many papers, this section will be brief or empty. If you only used AI for writing assistance and not for the research itself, that is perfectly fine. The card adapts to your actual usage.

3. Writing and Presentation

This is often the most relevant section for researchers. It captures AI involvement in the production of the paper itself. The section covers text generation, paraphrasing, translation, summarization, figure creation, table formatting, slide preparation, and any other presentation tasks where AI tools contributed.

Being specific here is important. "ChatGPT assisted with writing" tells a reviewer very little. "ChatGPT (GPT-4) was used to generate initial drafts of the related work section, which were then substantially revised and verified against primary sources" gives reviewers the context they need to evaluate your work fairly.

This section also asks about the degree of human oversight applied to AI-generated content. Did you review every output line by line? Did you use AI suggestions as a starting point and rewrite extensively? Or did you incorporate AI-generated text with minimal changes? Honest answers here build trust with your readers.

4. Ethics and Limitations

The final section asks you to reflect on the ethical dimensions of your AI usage. This includes any steps you took to mitigate bias in AI outputs, how you handled errors or hallucinations, and what limitations the AI involvement introduces to your work.

If your AI tool was trained on data that might underrepresent certain populations, and you used that tool for literature review or data analysis, that is worth noting. If the AI model you used has known failure modes that could affect your results, readers deserve to know. This section encourages the kind of critical self-reflection that strengthens research.

It also covers data privacy considerations. If you uploaded research data to a cloud-based AI service, did you consider whether that data contained sensitive information? Did you check the AI provider's data retention policies? These are real concerns that the ethics section prompts you to address.

How to Create an AI Usage Card

You have two main options.

The online generator at ai-cards.org. This is the fastest approach. You fill in the fields through a guided interface, and the tool generates a formatted card you can download as a PDF and include in your paper as a supplementary document or appendix. It is free and does not require an account.

The LaTeX template on Overleaf. If you are writing your paper in LaTeX and want to integrate the AI Usage Card directly into your document, the Overleaf template gives you a clean, professional format that matches common conference and journal styles. This is popular with researchers submitting to venues like ACL, NeurIPS, or JCDL.

Both options produce the same standardized format, so use whichever fits your workflow.

Who Is Using AI Usage Cards

Since the framework was published at JCDL 2023, adoption has grown steadily. Individual researchers were the earliest adopters, attaching AI Usage Cards to their preprints on arXiv and submitted manuscripts. Several conferences have since begun encouraging or requiring AI usage documentation, and AI Usage Cards provide a ready-made format that satisfies those requirements.

University writing centers and graduate programs have also started recommending the framework. For thesis students who need to document their AI tool usage (a growing requirement at many institutions), AI Usage Cards offer a structure that is more thorough than a simple declaration statement. If you are writing a thesis, our guide on how to disclose AI usage in your thesis covers this in detail.

How AI Usage Cards Relate to Other Frameworks

AI Usage Cards are specifically designed for documenting AI usage in research projects. They are different from Model Cards (which document AI models themselves), Datasheets for Datasets (which document training data), and System Cards (which document deployed AI systems). These frameworks are complementary, not competing.

If you are curious about how they compare, see our pages on AI Usage Cards vs Model Cards and AI Usage Cards vs Datasheets.

The Bottom Line

AI Usage Cards exist because AI is now part of the research process for millions of scientists, and the scientific community needs a consistent way to talk about that. They are not about policing AI use or making researchers jump through hoops. They are about giving you a simple, standardized way to be transparent about your workflow, so your readers and reviewers can evaluate your work with full context.

Creating one takes about ten minutes. The benefits to your credibility and your readers' ability to assess your work last far longer. Generate your free AI Usage Card now.